PDF Merge & Split

Because a PDF is not plain text — it is an object graph plus a cross-reference table. Objects reference each other by number: object 3 might be a page, object 5 its content stream. Each document has its own numbering scheme, so a naive concatenation produces conflicting numbers and a cross-reference table pointing at the wrong offsets. The result is a file that no reader will open. The correct approach is to parse each page’s objects, assign fresh numbers, and build a new cross-reference table — exactly what this tool does.

Three categories are explicitly unsupported: encrypted or password-protected files; files using object streams (where multiple objects are compressed into a bundle); and files using cross-reference streams. The latter two are compressed structures introduced after PDF 1.5 and are common in output from newer generators. In these cases the tool says so plainly rather than guessing — because a failed parse easily yields a file whose structure looks plausible but which will not open, and you would typically only discover that after sending it to someone.

Usually it is close to the sum of the inputs, which is expected: the page content itself has to be stored. It can be slightly smaller because redundant metadata is dropped, or slightly larger because a new cross-reference table and page tree had to be written. There is theoretical room for savings by sharing fonts or images referenced from multiple pages, but that requires far more complex resource restructuring; this version prioritises correctness over such optimisation.

Only the pages you selected. This deserves emphasis because it is an easy implementation trap: every page object records which parent node it belongs to, and that parent typically holds every page in the document. If you copy the parent along for convenience, the output silently includes pages you never selected — you think you shared page 2 while the full contents of page 1 are in the same file. This tool skips that parent reference when assembling a new document and attaches only your selected pages to a fresh page tree, with a dedicated regression test guarding the property.

Separate groups with commas; each group is either a single page (5) or a range (1-3). For example 1-3,5,8-10 produces three files: pages 1 to 3, page 5, and pages 8 to 10. Anything beyond the page count is ignored automatically, and reversed ranges such as 5-3 are skipped as invalid. Being permissive is deliberate — page numbers copied from elsewhere often bring extra digits, and failing hard would be more annoying than helpful.

Yes. In merge mode you can arrange the order of the files. If you also need to reorder pages within a file, split it into single pages first and then merge them back in the sequence you want.

Yes. Every output is a structurally complete, self-contained PDF with its own catalog, page tree and cross-reference table. It opens, prints and sends independently and does not need the original file.

No. The entire parsing and reassembly happens inside your browser; no step sends PDF data over the network. That is the main reason for writing the parser ourselves: this tool deals with contracts, ID documents and receipts, and if users cannot verify that a third-party library stays offline, that suspicion alone is enough to stop them using it.